Hakim

Speech-to-text service implementation using Hakim’s Arabic-first realtime transcription API

Overview

HakimSTTService transcribes audio into text using Hakim’s Arabic-first realtime speech-to-text API. It is a WebsocketSTTService that opens one persistent connection to WSS /v1/audio/transcriptions/stream, forwarding audio continuously and relying on Pipecat’s own VAD to signal end-of-turn — the same pattern used by other continuous-streaming STT plugins in this framework (e.g. AssemblyAI, Deepgram).

Source Repository

Source code, examples, and issues for the Hakim integration

Hakim Docs

Learn more about Hakim’s speech services

API Keys

Create and manage your Hakim API keys from the dashboard (Settings -> API keys)

Installation

This is a community-maintained package distributed separately from pipecat-ai, built and maintained by the Hakim team. It is not yet published to PyPI, so install it directly from GitHub:

uv pip install git+https://github.com/tryHakimAI/hakim-pipecat.git

Prerequisites

Hakim Account Setup

Before using the Hakim STT service, you need:

1. Hakim Account: Sign up at tryhakim.ai
2. API Key: Generate a key (format: hk_live_...) from the dashboard (Settings -> API keys), with the stt:write scope

Required Environment Variables

• HAKIM_API_KEY: Your Hakim API key. Falls back to this environment variable if api_key is not passed to the constructor.

Configuration

api_key · str · default: "None"

Hakim API key. Falls back to the HAKIM_API_KEY environment variable if not provided.

language · Language · default: "Language.AR"

Language hint for transcription.

input_audio_format · str · default: "pcm16"

Audio format of the input stream. One of "pcm16", "opus", or "mulaw". Pipecat’s transport delivers PCM16 by default, so the default matches without any client-side transcoding.

sample_rate · int · default: "16000"

Must match the sample rate of audio frames actually sent to the service.

region · str · default: "auto"

Pins the session to a specific regional endpoint. One of "auto", "de", "uae", or "ksa". "auto" picks the closest healthy region.

base_url · str · default: "None"

Overrides the region table entirely (staging / self-hosted). Takes precedence over region.

settings · HakimSTTService.Settings · default: "None"

Runtime-updatable settings. See Settings below.

Settings

Runtime-updatable configuration passed via the settings constructor argument using HakimSTTService.Settings(...).

Parameter	Type	Default	Description
timestamps	str	"segment"	One of "word", "segment", or "none".
diarize	bool	False	Speaker diarization. Only useful for stereo audio with one speaker per channel (e.g. call recordings) — mono diarize=True transcribes without speaker labels.
partials	bool	True	Whether to emit interim (non-final) transcripts.

See the source repository for the authoritative, up-to-date configuration options.

Usage

import os
from pipecat.pipeline.pipeline import Pipeline
from pipecat.transcriptions.language import Language
from hakim_pipecat import HakimSTTService

stt = HakimSTTService(
    api_key=os.getenv("HAKIM_API_KEY"),
    language=Language.AR,
)

pipeline = Pipeline([
    transport.input(),               # audio/user input
    stt,                              # Hakim STT transcription
    context_aggregator.user(),        # add user text to context
    llm,                              # LLM generates response
    tts,                               # text to speech synthesis
    transport.output(),               # stream audio back to user
    context_aggregator.assistant(),   # store assistant response
])

Compatibility

Tested with Pipecat v1.5.0. Check the source repository for the latest tested version and changelog.